Pull latest code
Fetch the newest commit. No change, no run.
From parsing logs to full-stack observability: practical tools built with Python, Streamlit, and deep JVM knowledge, augmented by lightweight AI integrations for performance engineering and incident management.
Every night, AI agents load-test a real app, self-heal their own script, flag regressions against past runs and trace errors to the exact line of code. Token-smart, serverless, and pluggable into your own app and LLM.
Fetch the newest commit. No change, no run.
One user, 20 s: is the script still valid?
A broken script is repaired by an agent.
250 users, 1 min; CPU, memory, network tracked.
Response times, errors and spikes vs earlier runs.
Stack traces and GC pauses vs the spikes.
Trace each problem to a file, line and cause.
Locust, infra and fix report on GitHub Pages.
Agent steps use an LLM, only when needed. The rest is plain code, and unchanged code is never re-tested.
At 250 users, JPetStore's checkout returned HTTP 500. Traced to OrderService.java: order ids come from a non-atomic read-then-increment, a race that only appears under load.
Sanity test passed (96 requests).
Implement robust unique ID generation for orders
Effort: highA few lines of config point the pipeline at your repo, your Locust journeys and any OpenAI-compatible model.
[target]
repo_url = "https://github.com/acme/shop.git"
build_cmd = "npm ci && npm run build"
start_cmd = "npm start"
base_url = "http://127.0.0.1:3000"
locustfile = "load/journeys.py"From local inference to cloud APIs and agentic frameworks.
MCP servers host lightweight training and inference pipelines close to telemetry, enabling continuous fine-tuning on project data and low-latency, context-aware inference. Use them to run anomaly detectors, score and prioritize alerts, synthesize concise RCA summaries, and surface remediation suggestions, all of which reduce manual forensics and speed MTTR.
Hover the pipeline steps for details. Open "More details" for the project layout and quickstart.
Goal: reduce time-to-diagnosis, make observability actionable, and ship reproducible developer tools with clear quickstarts, dashboards, and integrations.